Skip to content

README.md

Latest commit

 

History

History
114 lines (76 loc) · 2.92 KB

README.md

File metadata and controls

114 lines (76 loc) · 2.92 KB

🍳 AI Cooking Assistant

A conversational AI-powered cooking assistant built with Streamlit. Ask it for recipes, modify them on the fly, or ask general cooking questions — all in a clean, dark-themed chat UI.

Features

  • Recipe generation — ask for any dish and get a structured recipe with ingredients, steps, cooking time, and difficulty
  • Recipe modification — follow up with constraints like "make it vegan" or "no garlic" and the recipe updates
  • General cooking Q&A — ask anything about cooking techniques, substitutions, or tips
  • Session memory — the assistant remembers the current dish and all your requests within a session
  • Sidebar history — see your active dish and recent requests at a glance

Project Structure

├── app.py              # Streamlit UI
├── main.py             # LLM calls, prompt builder, validation logic
├── pydanticLayer.py    # Pydantic models and recipe validation
├── .env                # API key (not committed)
└── requirements.txt    # Python dependencies

Setup

1. Clone the repo

git clone <your-repo-url>
cd ai-cooking-assistant

2. Install dependencies

pip install -r requirements.txt

3. Add your API key

Create a .env file in the project root:

API_KEY=your_openrouter_api_key_here

Get a free API key at openrouter.ai.

4. Run the app

streamlit run app.py

Requirements

streamlit
requests
pydantic
python-dotenv

How It Works

  1. User input is sent to build_prompt() in main.py, which includes the current dish, session constraints, and the last known recipe as context.
  2. The prompt is sent to the LLM via the OpenRouter API (main.py → call_llm()).
  3. The response is parsed — if JSON is detected, it's extracted and validated against the Pydantic Recipe schema (pydanticLayer.py). If it's plain text, it's returned as a conversational reply.
  4. On validation failure, the LLM is retried up to 2 times with the error message included in the prompt.
  5. The result is rendered in the Streamlit UI — recipes as styled cards, text as chat bubbles.

Configuration

Variable Location Description
API_KEY .env Your OpenRouter API key
MODEL main.py LLM model to use (default: meta-llama/llama-3.1-70b-instruct)
max_retries main.py Number of retry attempts on invalid JSON (default: 2)

To switch models, update the MODEL variable in main.py:

MODEL = "meta-llama/llama-3.1-70b-instruct"  # recommended
# MODEL = "openai/gpt-4o-mini"               # alternative

Note: Smaller models (e.g. llama-3-8b-instruct) may produce unreliable JSON. A 70B+ model is recommended for best results.

CLI Mode

You can also run the assistant in the terminal without Streamlit:

python main.py

License

MIT